Thursday, February 03, 2005

Book Review: On Intelligence

This is a book that the publisher describes thus: “How a new understanding of the brain will lead to the creation of truly intelligent machines.” The description is truly misleading because it omits the fact that this is really a thoroughly enjoyable romp through the human brain.

After all, how much enjoyment can you reasonably expect from a book about brains? Education? Yes. Enlightenment? Yes. But enjoyment? Not much unless that author, who is known best for helping start Palm computing and Handspring and for being the architect of devices such as the Palm Pilot and Treo, makes declarations such as “I am crazy about brains” and “…I have to admit I am a neocortex chauvinist.”

The author set out to answer a number of major questions that have floated around the computer and artificial intelligence communities for years:

  • Can computers be intelligent?
  • Weren’t neural networks support to lead to intelligent machines?
  • Why has it been so hard to figure out how the brain works?
  • What is intelligence if not defined by behavior?
  • How does the brain work?
  • Can we build intelligent machines and what will they do?

In addressing these questions, and others, the author sets out a new theory of intelligence and how the brain works in a manner that is clearly understandable. A highlight is the author’s discussion of the neocortex, the brain’s seat of intelligence. This critical part of the brain is described in a manner that is quite understandable to the reader who lacks medical training. The examples are real world, often based on common themes that touch everyone.

The computer industry has long suggested that once there is a magic combination of process speed, memory size and near instant retrieval times, then it will be able to produce a viable intelligent machine that can mimic the human brain. This theory is soundly put to rest as the author explains the fundamental differences between a digital (or analog) computer and the human brain. He clearly differentiates between the creation of an intelligent machine that can do certain things exceedingly well, and the creation of a device similar to the human brain. He is unequivocal in his position, stating, “I don’t think we will build intelligent machines that act like humans, or even interact with us in humanlike ways.” There are, he notes, no R2D2, C3PO or HAL’s in our future – immediate or otherwise.

Limited applications robots, such as smart cars and appliances are definitely in our future, he contends, but is understandably reluctant to get overly predictive in terms of details. Nor does he predict the ultimate uses for intelligent machines. This, however, is not a cop-out, as he explains the basis for his reluctance and describes the advances in technology (speed, capacity, replicability, and sensory systems) that are necessary to produce future intelligent devices.

The author completes with work with a thoughtful list of 11 testable predictions that one can use to track the growth and process of machine intelligence and the future development of intelligent machines.

Jeff Hawkins may be a fan of the human brain, but I have become a fan of Jeff Hawkins.

On Intelligence

By Jeff Hawkins, with Sandra Blakeslee
ISBN: 0-8050-7456-2
Published by: Times Books, Henry Holt and Company, LLC
260 pages.


No comments: